Conversation intelligence API: build call analytics on Bitrix24 REST or use a ready API?

· By , Founder of Aelo

A conversation intelligence API takes a call recording or a chat transcript and returns structured analysis — a quality score, a summary, scores for each part of the conversation, sale probability — rather than raw text. It sits one layer above a speech-to-text API: transcription, scoring and storage happen on the vendor’s side, and you read back the result.

Aelo is mentioned throughout because we built it. The Bitrix24 part describes the platform’s own REST API and applies whatever you choose.

What is a conversation intelligence API?

It is the layer that turns a conversation into numbers and text a CRM or a report can use. Three kinds of API get called “call analytics”, and they return very different things:

APIYou get backStill yours to build
Speech-to-text APIA transcriptScoring, rubric, storage, CRM write-back
General-purpose LLM APIWhatever the prompt asks forTranscription, rubric that stays stable, storage
Conversation intelligence APIA record and its analysis, in a fixed contractOnly the integration with your systems

The difference that matters in production is the fixed contract: the same documented fields for every call, and a score that is the one the product itself shows, so the API, the dashboard and your reports agree. With your own prompt, keeping the output shape and the scoring stable across prompt edits and model changes is your job.

What does it take to build Bitrix24 call analytics on the REST API yourself?

The steps below are the end-to-end path Aelo’s own Bitrix24 app takes. Steps 1, 3, 7 and 8 call the Bitrix24 REST API; steps 2 and 4–6 are your own receiver and analysis pipeline; step 8 matters only if you want calls from before the integration.

  1. Learn that a call ended. Register an app (OAuth 2.0) and subscribe with event.bind to ONCRMACTIVITYADD and ONCRMACTIVITYUPDATE. An outgoing call usually arrives already completed on ADD; an incoming one turns COMPLETED to Y on UPDATE. The same call can therefore arrive twice — you need an idempotency key.
  2. Acknowledge fast. Repeated error responses make Bitrix24 deprioritize and eventually unbind your handler, which silently stops delivery for that portal. The receiver should return 200 at once and put the work on a queue.
  3. Reach the recording. crm.activity.get gives you the activity’s FILES; disk.file.get turns a file id into a DOWNLOAD_URL. The recording is often attached some time after the call completes, so retry for a while before deciding a call has none. Download and store the audio yourself.
  4. Transcribe it with a speech-to-text API of your choice, and handle calls with no recording, missed calls and very short ones.
  5. Redact personal data before the text goes to a language model — at least phone numbers, emails and payment details.
  6. Score it. Write the rubric, the prompt and the parsing, and keep scores comparable from one month to the next.
  7. Write the result back with crm.timeline.comment.add on the CRM record the call belongs to, or attach the transcript to the call with telephony.call.attachTranscription.
  8. Cover the past. Calls made before the integration existed come from crm.activity.list filtered to call activities (PROVIDER_TYPE_ID: 'CALL'), not from events.

None of these steps is hard on its own. The cost is keeping all eight working: token refresh, retries, a portal that reinstalls the app, a rubric the sales team wants to change.

When does building it yourself make sense?

Build it yourself when the analysis is your product: you need full control over models, prompts and where every byte is stored, or you sell analytics to others. Use a ready conversation intelligence API when you need scores in your CRM and reports for a sales team, and the pipeline is a cost, not the point.

What does Aelo’s public API return?

All endpoints live under https://api.aelo.cloud/api/public/v1 and take a bearer key (aelo_sk_…). Only the organization owner can issue keys, in Settings → API Keys; each key carries one or both scopes, records:write and records:read, and every request is limited to the organization that issued the key.

EndpointWhat it does
POST /recordsCreate a record from a fileUrl (calls) or textContent (chats)
POST /records/uploadCreate a call record from an uploaded audio file
GET /recordsList records, newest first
GET /records/{recordId}Get one record
GET /records/{recordId}/analysisGet the latest analysis for a record
  • Analysis. A summary, qualityScore — the same number the dashboard shows for that call — six rubric scores (greeting, needs, clarity, objections, closing, politeness) and saleProbability for calls. A metric the analysis did not produce is null, never a 0.
  • Idempotency. Send your own externalId, and a retry returns the original record instead of creating a second one.
  • Polling. While a record has no analysis yet, analysis_not_found carries details.analysisExpected: true means an analysis is still coming — in Manual mode, only once someone starts it in Aelo — and false means the record finished without one — for example a call below the project’s minimum length.
  • Limits. 200 requests per minute per key on each endpoint; file-size limits for uploads are in the limits table.
  • Plan. Starter and above.

The full reference is in the API documentation, and both upload paths are covered in Uploading calls and chats.

Do you need the API if your calls are already in Bitrix24?

Usually not. The Aelo app for Bitrix24 does steps 1–7 above without code: once Automatic mode is on, it picks up finished calls and Open Channel chats, analyzes them and writes the report to the timeline of the CRM record the call or chat belongs to. Conversations from before the installation are not picked up on their own: the organization owner imports a chosen period. The public API is for everything else — your own telephony, a data pipeline, or a CRM Aelo does not integrate with yet. For how the native app compares with Bitrix24’s built-in AI, see Aelo vs. BitrixGPT.

Frequently asked questions

Is Aelo a conversation intelligence API?

Yes. Aelo has a public REST API at https://api.aelo.cloud/api/public/v1. You send a call recording (a file or a URL) or a chat transcript and read back the record and its analysis: a summary, a quality score — the same number the dashboard shows — six rubric scores and, for calls, sale probability. Keys are org-scoped bearer tokens with records:write and records:read scopes. It is available on Starter and above.

Can I get the call transcript through the Aelo API?

Not today. The public API returns records and their analysis — summary, quality score, rubric scores and sale probability — but has no transcript endpoint. The transcript stays in Aelo, and in the Bitrix24 app it is also attached to the call itself. If your pipeline needs raw text, a speech-to-text API is the right layer for that part.

How much does the Aelo API cost?

The public API is included on Starter and above; public paid plans start at $59 a month and are priced on analyzed minutes, not per API call. Free-plan organizations cannot issue keys. When the included minutes and the credit balance both run out, a new upload is refused with insufficient_balance, and that request stores nothing.

Which Bitrix24 REST methods does a DIY call analytics pipeline need?

At minimum: event.bind for ONCRMACTIVITYADD and ONCRMACTIVITYUPDATE to learn that a call ended, crm.activity.get and disk.file.get to reach the recording, crm.activity.list for past calls, and crm.timeline.comment.add or telephony.call.attachTranscription to write results back. Transcription, scoring and storage are outside Bitrix24 — you build or buy those.

Can I send chats to the API, not only calls?

Yes. POST /records takes a chat as textContent instead of a fileUrl. A very long chat is stored but not analyzed; the exact size limits are in the API reference. A chat with too few messages or no operator reply finishes without an analysis, and the API says so instead of leaving you polling.